Frame an ENA Study: Units, Conversations, and Codes
Translate a relational research question into aligned units of analysis, conversation boundaries, and theory-linked codes before opening the software.
Le corps du tutoriel vérifié est actuellement disponible en anglais.
Sources méthodologiques
Tutoriel complet
ENA is useful when the organization of connections matters, not only the frequency of individual ideas or actions. The method models co-occurrences among codes within a defined context, accumulates those connections for each unit, and places the resulting networks in a shared space for comparison.
The most consequential choices therefore happen before a plot appears. A research question, unit of analysis, conversation boundary, segmentation rule, and coding scheme should form one coherent design. If those pieces answer different questions, the software may still produce a network, but its meaning will be unstable.
Cas pédagogique
Do scaffolds change how design ideas connect?
A fictional study asks how teacher-design teams connect goals, student evidence, strategies, tradeoffs, and revisions during a feedback-activity review. Four teams use a reflection scaffold and four do not. The dataset is intentionally small and synthetic so the pathway can teach model logic without presenting invented findings as research evidence.
Jeu de données d'exercice
Étape 1
Write a relational question
Ask about the structure of connections: How do scaffolded and baseline teams differ in how they connect student evidence with proposed strategies and revisions? This is more aligned with ENA than asking only which group mentions evidence more often. State why co-occurrence within interaction context represents a meaningful relation for the phenomenon.
Point de contrôleThe question names a relational pattern, a comparison, and the context in which proximity is meaningful.Étape 2
Choose the unit of analysis
The unit identifies the network that will be accumulated and compared. In the teaching study, each team is one unit because the question concerns team reasoning. Choosing individual speakers instead would answer a different question and would require an account of how shared talk is assigned to people.
Point de contrôleYou can complete the sentence: one network represents all coded connections for one team.Étape 3
Bound the conversation
A conversation prevents connections from crossing contexts that should remain separate. Here, each design-review round is a conversation within a team. For moving-window accumulation, line order also matters. The boundary and window should reflect how long ideas remain meaningfully connected in this activity, not a convenient software default.
Point de contrôleThe design states where a conversation starts and ends and why links should not cross that boundary.Étape 4
Define theory-linked codes
Use a small fixed set that represents the constructs in the question: Goal, Evidence, Strategy, Tradeoff, and Revision. Define inclusion and exclusion rules, examples, segmentation, and whether multiple codes can occur on one row. Codes are analytical claims about evidence, not neutral labels generated by the plot.
Point de contrôleEach code has a definition, a counterexample, and a reason it belongs in the relational model.Étape 5
Specify the comparison and metadata
Condition is comparison metadata, not a code. Decide in advance which units belong to baseline and scaffolded groups, which contextual fields should be retained, and which contrasts are exploratory. In a real study, plan sample size and statistical strategy before interpreting group distance.
Point de contrôleGrouping fields are separate from the codes whose connections form the networks.Étape 6
Write the model specification before analysis
Create a one-page record containing the question, unit, conversation fields, row order, code columns, window rationale, weighting, normalization, rotation, planned groups, exclusions, and sensitivity checks. Mark decisions made before viewing results. This record turns the model from a collection of clicks into an auditable analysis.
Point de contrôleA second analyst could configure the intended model from the specification alone.